VALL-E

Training compute

10¹⁹ FLOP

Parameters

353M

Published

Jan 5, 2023

VALL-E is an AI model developed by Microsoft (United States), first published in January 2023. It works in the audio and speech domain, on tasks such as speech synthesis and text-to-speech (tts).

Training it took an estimated 10¹⁹ FLOP of compute (estimation method: operation counting). The model has 353,000,000 parameters. It was trained on roughly 76.8B datapoints. Training ran on NVIDIA V100. The compute alone is estimated at $11 in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,180 citations. Epoch AI rates the confidence of this record as speculative.

Full record

Organization: Microsoft
Country of organization: United States
Domain: Audio, Speech
Task: Speech synthesis, Text-to-speech (TTS)
Training compute: 10¹⁹ FLOP
Compute estimation method: Operation counting
Parameters: 353,000,000
Dataset size: 76.8B
Training hardware: NVIDIA V100
Training cost (2023 USD): $11
Model accessibility: Unreleased
Open weights: No
Citations: 1,180
Epoch confidence: Speculative